fix(neural): the learning cycle crashed on every run — retune and first real training

config/config.exs

@@ -1,12 +1,14 @@
 # SPDX-License-Identifier: MPL-2.0
 import Config
 
-# verisim-data: the canonical git-backed flat-file store (300 scans, 23 recipes)
-# Lives inside the nextgen-databases monorepo
+# verisim-data: the canonical git-backed flat-file store (300 scans, 23 recipes).
+# No machine-specific defaults (same migration as #918): the in-repo store ships
+# as `data/verisim` and deployments point elsewhere via HYPATIA_VERISIM_DATA_PATH
+# / HYPATIA_FLEET_PATH. The paths are expanded relative to the hypatia repo
+# root, so run the app from there (the documented `just`/`mix` workflow does).
 config :hypatia,
-  verisimdb_data_path:
-    Path.expand("~/Documents/hyperpolymath-repos/nextgen-databases/verisim/verisim-data"),
-  fleet_path: Path.expand("~/Documents/hyperpolymath-repos/gitbot-fleet"),
+  verisimdb_data_path: System.get_env("HYPATIA_VERISIM_DATA_PATH", "data/verisim"),
+  fleet_path: System.get_env("HYPATIA_FLEET_PATH", "../gitbot-fleet"),
   http_port: 9090
 
 # Import environment specific config (config/test.exs, config/dev.exs, etc.)

data/verisim/health/hypatia.json

@@ -1,37 +1,35 @@
 {
-  "checked_at": "2026-03-06T15:12:12.436980Z",
+  "checked_at": "2026-10-15T17:28:15.241699Z",
   "checks": {
     "circuit_breaker": {
       "failures": 0,
       "state": "closed"
     },
     "filesystem": {
-      "missing": [
-        "data/verisim/dispatch"
-      ],
-      "status": "fail"
+      "paths_checked": 5,
+      "status": "pass"
     },
     "fleet_access": {
-      "last_modified": "{{2026, 3, 6}, {15, 12, 2}}",
-      "size_bytes": 945295,
+      "last_modified": "{{2026, 10, 13}, {23, 32, 54}}",
+      "size_bytes": 132,
       "status": "pass"
     },
     "learning_scheduler": {
-      "confidence_updates": 0,
-      "last_run": "~U[2026-03-06 15:12:02.928002Z]",
-      "outcomes_processed": 6914,
+      "confidence_updates": 2,
+      "last_run": "~U[2026-10-15 17:28:10.243987Z]",
+      "outcomes_processed": 6964,
       "status": "pass"
     },
     "recipe_health": {
       "low_confidence": [],
       "status": "pass",
-      "total_recipes": 5
+      "total_recipes": 66
     },
     "verisimdb_data": {
       "outcome_files": 2,
-      "recipe_files": 5,
+      "recipe_files": 66,
       "status": "pass"
     }
   },
-  "overall": "degraded"
+  "overall": "healthy"
 }

data/verisim/neural-states/esn.json

@@ -0,0 +1,10 @@
+{
+  "accuracy": {
+    "mse": null,
+    "samples": 0,
+    "mae": null
+  },
+  "drift": "insufficient_data",
+  "predictions_count": 0,
+  "trained": true
+}

data/verisim/neural-states/gnn.json

@@ -0,0 +1,8 @@
+{
+  "clusters": {},
+  "edge_weights": [],
+  "embedding_dim": 16,
+  "layer_count": 3,
+  "node_embeddings": {},
+  "trained": false
+}

data/verisim/neural-states/lsm.json

@@ -0,0 +1,11 @@
+{
+  "anomaly_count": 0,
+  "history_length": 0,
+  "summary": {
+    "anomaly_count": 0,
+    "current_activation": 0.0,
+    "history_length": 0,
+    "recent_trend": "stable"
+  },
+  "trained": false
+}

data/verisim/neural-states/moe.json

@@ -0,0 +1,156 @@
+{
+  "expert_stats": [
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "documentation",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "accessibility",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "infrastructure",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "performance",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "quality",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "security",
+      "total": 0
+    },
+    {
+      "activations": 0,
+      "correct": 0,
+      "domain": "sustainability",
+      "total": 0
+    }
+  ],
+  "expert_weights": [
+    {
+      "bias": 0.5,
+      "domain": "documentation",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "accessibility",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "infrastructure",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "performance",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "quality",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "security",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    },
+    {
+      "bias": 0.5,
+      "domain": "sustainability",
+      "weights": {
+        "severity": 0.3,
+        "complexity": 0.2,
+        "fix_rate": 0.15,
+        "frequency": 0.2,
+        "recency": 0.15
+      }
+    }
+  ],
+  "gating_weights": [
+    {
+      "domain": "documentation",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "accessibility",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "infrastructure",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "performance",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "quality",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "security",
+      "weight": 0.14285714285714285
+    },
+    {
+      "domain": "sustainability",
+      "weight": 0.14285714285714285
+    }
+  ],
+  "last_trained": null
+}